Improve the quality of error messages emitted by Weaver

[lquerel]

The deserialization of YAML semconv files using serde often results in low-quality error messages.

This PR:

• Continues to use serde for deserialization.
• Implements a fallback mechanism when serde detects errors:
  • Generates the JSON schema associated with the Rust SemConvSpec structure.
  • Converts the YAML file into an equivalent JSON representation.
  • Validates the JSON content obtained from the previous step using the generated JSON schema.
  • Exploits error messages returned by the validator (if any).
  • For each error, identifies its original location in the YAML file, extracts the description field from the JSON schema, and generates detailed error messages using the miette framework.

This PR also adds a new parameter to the command weaver registry json-schema to generate the JSON Schema for the semconv-group instead of the resolved registry.

weaver registry json-schema -j semconv-group

Example of error messages:

Closes: #746

[jsuereth]

I like how this forces better rust documentation too.

[lquerel]

Yes, I’ve improved a few comments, but we’ll certainly need to continue this effort throughout the objects representing all the semantic conventions.

[jsuereth]

It looks like just generally more investment in Miette would be valuable from what I see.

E.g. we should see if we can preserve spans further into resolution for error messages over time.

[jerbly]

I was going to say something similar I think - these errors are now better than the ones produced here: https://github.com/open-telemetry/weaver/blob/main/crates/weaver_semconv/src/group.rs#L107.

On the topic of using this machine generated json schema. The hand-crafted json schema: https://github.com/open-telemetry/weaver/blob/main/schemas/semconv.schema.json encodes more rules than can be expressed in the Rust structs/enum, I think. And it is aligned with --future as @lmolkova pointed out. So it would be a shame to lose this. An author with a modern text editor with the schema plugged in will be warned there and then that prefix is not part of the schema. With the machine generated version there won't be this warning until they pass it through weaver. This IMO creates a disconnected experience for the author where they'll have to go back and forth between command line and text to make their yaml compliant.